Teepan.

Building and maintaining backend services and APIs (REST and GraphQL) across the Radio France digital platform — handling high-concurrency traffic with strict uptime requirements. Comfortable across the full stack: from API design to production front-end on the public website and internal tools.

• Developed and maintained multiple backend services and APIs (REST and GraphQL) across the digital platform, handling high-concurrency traffic with strict uptime requirements.
• Built and maintained the BFF API (Node.js, Fastify, Apollo Server) — data orchestration layer for the Radio France mobile app.
• Optimized PostgreSQL query performance under high concurrency
• Built full-stack features on radiofrance.fr (React/Symfony) and internal back-office tools (Svelte/SvelteKit).
• Deployed and monitored services in Kubernetes with Docker and CI/CD pipelines (GitLab CI).
• Integrated AI tooling into the dev workflow: built a Jira-to-Claude Code automation pipeline for ticket triage and code generation.

AI product that lets users upload conversation exports (WhatsApp, iMessage) to converse with an AI that reproduces a specific person's communication style and relational dynamic. Unlike generic chatbots, Revive captures how two specific people talk to each other: tone, humor, patterns, inside references.

• Designed a Persona Extraction + RAG + Dynamic System Prompt architecture that analyzes conversation history to build a behavioral profile of the target person.
• Built a semantic search layer using PostgreSQL with pgvector. Conversation fragments are embedded, indexed, and retrieved by relevance to generate contextually accurate responses.
• Integrated Claude API (Anthropic) for response generation with structured prompt templates that inject persona traits, relational context, and retrieved conversation snippets.
• Full-stack implementation: Next.js front-end, Node.js/Fastify/TypeScript backend, Drizzle ORM, with structured conversation parsing and embedding generation.